Semantic AI Tools Help Inventors Generate Ideas, Patents

This is a sponsored article brought to you by IP.com.

Prospecting for technological opportunities, patenting novel inventions, and assessing a company's competitive position require the ability to match in-house knowledge and strengths (often described in idiosyncratic in-house terminology) with world-wide technical and patent literature.

Semantic artificial intelligence-working on hundreds of millions of pages of patent and technical articles-is making the process faster and easier than it has ever been.

IP.com was founded in 2008 to help innovators master the patent and technical literature and bridge the yawning gap between technology conception and commercialization-while also jumping the communication barriers that can cut R&D off from corporate legal, intellectual property, licensing, strategy, and marketing teams.

Semantic artificial intelligence tools, like the Semantic Gist engine used in IP.com's InnovationQ Plus, can help R&D engineers refine their ideas, find areas of opportunity, evaluate competing technologies, and greatly accelerate the patent-writing process.

IP.com makes a suite of ideation and analysis tools that help inventors refine their ideas and produce more-valuable patent applications and produce them sooner with fewer headaches. The cornerstones of the system are the proprietary Semantic Gist artificial intelligence and databases that include not only the worldwide patent literature (including the U.S., E.U., Japan, Korea, and others), but crucially, they also search the contents of papers in research databases.

Most notably, the company enjoys exclusive access to the full text of IEEE Xplore, arguably the world's most comprehensive collection of electrical and electronic engineering research publications, containing more than 5 million articles from more than 300 peer-reviewed journals, with contents ranging from today back to 1884, and some 25,000 new journal articles and conference papers every month.

Attendees at How to Make Sure Your Ideas Are Viable and Valuable," [1] IP.com's recent IEEE Spectrum Tech Insider webinar, were asked about their patenting ambitions. Given the topic, it is not surprising that 78 percent either had been named inventors or wanted to be. Twenty-nine percent said they already used ideation tools to help them set technical goals on the job, and 47 percent said that they did not yet, but would like to have access.

IP.com's web-based tools are designed to cover the intellectual-property-development process from initial ideation (IQ Ideas Plus and InnovationQ Plus) through development (InnovationQ Plus and IP Analytics) and into commercialization.

• IQ Ideas Plus, an AI-powered ideation tool that finds relevant prior art and sparks new ideas for consideration. The tool applies the theory of inventive problem solving" to codify challenges, beginning with either a) a description of the system and an outline of what you are trying to improve, or b) a description of a problem and an initial assessment of what failed. The tool helps the designer state the case simply and include up to five important considerations. These feed the semantic engine, which helps the developer search relevant technology and refine the invention concept. IQ Ideas Plus is thus a complete innovation management solution that helps bring novel ideas to market faster.
• InnovationQ Plus (the focus of this article) ferrets out comparable technologies, using AI-driven natural-language searching to put papers and patents in an international context and refine the competitive product- or technology-development process.
• IP Analytics take a fully described patent or technology, compares it against the patent and engineering literature, to assess what IP.com calls the vitality" of a patent or technology-including its uniqueness, defensibility, and competitive position. IP analytics can also assess the competitive strength of entire technology and patent portfolios.

Semantic AI

IP.com's Semantic Gist AI search engine learns from experience to associate verbal concepts and map them. It relies on, among other methods, on dimension reduction," or simplifying complex relationships so that meaningful patterns are easier to discern.

To oversimplify, constructing a natural-language AI like Semantic Gist requires five principal steps: build a knowledge graph, reduce dimensionality, train, predict, and confirm [2].

Build a knowledge graph. This step involves constructing a multidimensional map of words and phrases. Then assign each a position relative to all of the other terms; this can be done manually or automated, or a combination of both. Distance and position give the terms their meanings. (Note: People have been trying to develop exhaustive categorizations of knowledge since the Third Century Greek philosopher Porphyry of Tyre. [3] Anyone who tries quickly finds that the apparently simple task is, in fact, fiendishly complex.)

Reduce dimensionality. Simplify relationships to make patterns easier to identify. A data-compression algorithm, for example, reduces the complexity of a text file to a smaller number of repeated elements. Or take the illustration (Figure 2, from Belgian artist Fred Eerdekens), which reduces the dimensionality of a seemingly meaningless tangle of wires to a 2-D projection of the word light."

Train. Machines learning proceeds by repetition and correction. The training can be assisted-a human user tells the AI when it has solved a test problem correctly-or unassisted-the machine keeps varying its solution until it approaches a correct result. Through trial and error, the AI digests the test input (such as patent filings or journal papers) and connects them with concepts in the knowledge base, assigning strengths or frequencies to the associations.

Predict. Once the algorithm is trained, it can be evaluated. Based on the dimension-reduced input data, the AI predicts how likely it is that reduced elements (the full text of two papers on low-power radio, for example) are related. The AI can then reconnect the representation with the higher-dimensional data-the original full text article--and suggest that they cover closely related topics.

Confirm. One of the delights of machine learning and artificial intelligence is their ability to reveal unexpected relationships that have previously been hidden in undecipherably complex data. But sometimes they make mistakes: Some quirk in coding and decoding the data representation leads to a faulty conclusion. A knowledgeable eye-coupled to what one engineer calls the Mark I processor," the brain-needs to review and confirm what the machine discovered.

Mapping Ideas

For example, given a search like wake-up radio AND low-power AND wave-form" limited to the past 10 years (showing the most recent work on waveforms best suited to ultra-low-power radios integrated into devices like battery-powered remote sensor to extend their service lives), InnovationQ Plus will take a second or two to rank and present the most relevant of more than 27 million patents and technical papers, and display them with text and graphics (Figure 3).

Engineers can page through results like this, or they can choose a variety of graphic presentations that help them both clarify their ideas and make new associations.

A semantic map of the same query (limited to non-patent literature this time) shows how key terms cluster, with cmos," lna," and baseband" closely related, but far away from ground" and earth" (Figure 4).

The same search on patents published during the same period, however, shows that the most prominent association is between instruction" and downlink," which are widely separated from reception" and wave"-and that Qualcomm is the leading filer in the market (Figure 5).

The ability to analyze patent or technical publications-or both together-helps the user track both research and commercialization.

The wake-up radio" illustration is oversimplified, of course. But the comparisons do begin to suggest strategies for finding where active research diverges from patent activity, potentially showing frontiers of opportunity.

The examples also show how useful it can be to reorganize information by a variety of criteria and present them in a number of forms, associating familiar and unfamiliar ideas.

Mapping Related Concepts

The Cooperative Patent Classification system (CPC) [4] is a sort of Dewey Decimal system for inventions.

The code consists of an initial Section letter (H for electricity" and G for physics" are the most common prefixes for electronic and electrical inventions), followed by a two-digit Class number, a single Subclass letter, a one- to three-digit Group number, a slash, and a two-digit (or more) Main Group number.

The U.S. Patent and Trademark Office and the European Patent Office developed the CPC jointly to the replace the similar International Patent Classification and their own earlier individual classification schemes.

For example, the first classification (a patent may have several) for US2524035 Three-electrode circuit element utilizing semiconductive materials" by John Bardeen and Walter H. Brattain (the patent for the transistor) is H03F 3/183.

IP.com parses this H03F 3/183 identifier like this:

InnovationQ Plus lets users sort and quantify patent-search results by both CPC and IPC codes.

In the process, users can both understand how patent offices will view their applications and, possibly, discover inventions whose relevance was not previously apparent.

Sorting the results for wake-up radio low-power wave-forms" by the CPC classifications that contain the largest number of relevant patents shows that the germane filings cluster in CPC codes H04J 13/12 and H04B 1/0082 (Figure 7).

Clicking on the codes translates them into English, or at least Patentese. Both fall into Class H04, which translates to Electricity: Electric Communication Technique. H04J 13/12 goes on to describe...

While H04B 1/0082 is...

It's worth noting, however, that some relevant inventions can be found in G06F-Electric Digital Data Processing for remote monitoring activity.

A handful even show up in AG1N-1/36014: Human Necessities / Medical or Veterinary Science; Hygiene / Electrotherapy; Magnetotherapy; Radiation Therapy; Ultrasound Therapy where they cover contact electrodes for electrical stimulation.

Mapping the Competitive Space

Consider a more specialized, and perhaps more realistic, question: Who is developing wave-forms for low-power transmission and reception in wake-up radio?"

Moving from the semantic map to a bar-graph representation shows that, while Qualcomm has published by far the greatest number of relevant patents over the past decade, Huawei has published more of the most relevant filings (shown in yellow).

We can break patent activity down geographically:

Geographic data can be displayed in pie charts or graphs, displaying patent collections, the inventor's country of origin, the applicant's country of origin.

InnovationQ Plus can even take a snapshot of overall patent enforceability.

We can even see who the most prolific individual inventors are (again, the most relevant patents are in yellow).

InnovationQ Plus offers more than 20 standard visualizations for patent queries and 9 for searches of non-patent literature. Semantic maps, term clouds, and term heat maps, and constellations of patent classification codes can help spur creativity.

Displays of activity by date (publication, priority, or expiration), by individual and corporate inventors, by geography (and more) help inventors understand their competitive position-including which opportunities might be blocked by others' IP, and which avenues may be open.

Beyond the standard visualizations, InnovationQ Plus offers custom charting, allowing, for example, a display of annual filing activity by the 10 most prolific companies.

Faster Communication, Faster Filings

Communication is the key to drafting patents faster. And efficient filing translates into earlier priority, which means market advantage.

But there are barriers. Any group of people working closely on a specialized tasks will develop its own ways of describing important problems, processes, and things. Research and development teams uncover new knowledge, solve new problems, develop new processes, and make new things, so their local language" can become very localized indeed.

While developing new intellectual property may spawn new language, patenting a novel, non-obvious invention demands retranslating the discovery into what can seem like a completely different tongue. Indeed, the corporate I.P. department, general legal team, marketers, and strategists may each seem to speak different dialects.

InnovationQ Plus offers more than 20 standard visualizations for patent queries and 9 for searches of non-patent literature, including semantic maps, term clouds, and term heat maps.

InnovationQ Plus (and, in fact, all IP.com tools) are structured into portfolios that can be saved and shared across R&D project teams, I.P. staff, and the rest of the organization. In the past, inventors might have assembled a raft of technical documents and an imperfect outline of the invention to forward to the patent department. A series of conversations with the patent attorney or agent might follow. Then, after a wait of days or weeks, the inventor would see a first draft that looked nothing like he or she had expected. A series of back-and-forth exchanges would follow before the inventor and I.P. group agreed on description, claims, illustrations, and citations.

The communications tools provided by IP.com shorten the process immensely. (This includes saving time on training: with a little instruction, a new user can begin producing worthwhile results in a few hours.) Inventors can home in on opportunities faster. They can search technical papers and prior art in a flash. They put together project portfolios that already include the citations and technical information the patent lawyers need.

And, indeed, the inventor and patent lawyers can be sure that patent offices are literally reading from the same page: InnovationQ Plus is used throughout the U.S. Patent and Trademark office as well as the Swedish Patent Office.

Along with Ideas Q Plus and IP.com's patent and technology vitality analyses, it all adds up to more efficient, more targeted, more productive R&D. Users report that introducing InnovationQ Plus substantially streamlined their patent-writing processes. Thus, over time, better, faster data analysis and better, faster internal communication add up to substantial competitive advantage.

Citations

[1] A Faster Way to Find and Validate New Ideas," IEEE Spectrum Tech Insider Webinar, 7 Oct. 2021. https://engineeringresources.spectrum.ieee.org/free/w_defa1630/prgm.cgi (accessed Dec. 09, 2021).

[2] Finding the Unexpected in Cell Signaling Studies | Biocompare: The Buyer's Guide for Life Scientists." https://www.biocompare.com/Editorial-Articles/364203-Finding-the-Unexpected-in-Cell-Signaling-Studies/ (accessed Dec. 09, 2021).

[3] J. F. Sowa, Semantic Networks," Accessed: Dec. 09, 2021. [Online]. Available: http://www.jfsowa.com/pubs/semnet.htm.

[4] P. Patel, M. Musen, and U. Y. Feroz Farazi, A semantic web primer," Accessed: Dec. 09, 2021. [Online]. Available: http://mitpress.mit.edu.